A Kalman Filter with a Perceptual Post-filter to Enhance Speech Degraded by Colored Noise

Speech enhancement algorithms have been employed successfully in many areas such as VoIP, automatic speech recognition and speaker verification. Some of the methods assume that the environmental noise is white noise. However, when used in colored noise environments, those methods will produce a weaker performance. Approaches for colored noise have also been previously proposed, however those previous methods have to detect non-speech frames for the noise covariance estimation. This paper proposes a method for colored noise speech enhancement based on a Kalman filter combined with a post-filter using masking properties of human auditory systems. No detection of non-speech frames is needed in the proposed method.